作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程

• 图形图像处理 • 上一篇    下一篇

结合运动特征的目标跟踪方法

周明珠,周治平   

  1. (江南大学物联网工程学院,江苏无锡214122)
  • 收稿日期:2014-04-08 出版日期:2015-03-15 发布日期:2015-03-13
  • 作者简介:周明珠(1988 - ),女,硕士,主研方向:图像处理,目标跟踪;周治平,教授、博士。
  • 基金资助:
    国家自然科学基金资助项目(61373126)。

Target Tracking Method Combined with Motion Feature

ZHOU Mingzhu,ZHOU Zhiping   

  1. (School of Internet of Things Engineering,Jiangnan University,Wuxi 214122,China)
  • Received:2014-04-08 Online:2015-03-15 Published:2015-03-13

摘要: 针对视频跟踪过程中出现的背景干扰、目标遮挡等问题,提出基于多特征融合的均值漂移算法和最小二乘 法轨迹预测跟踪方法。为解决背景干扰问题,使用改进的混合高斯模型对背景实施建模,提取运动前景目标,采用 提取出的运动信息结合颜色、纹理特征对目标进行描述,在跟踪过程中利用运动信息去除背景噪声的干扰,从而适应背景和目标的变化,得到目标位置,当遮挡发生时,根据目标遮挡前的先验信息预测最小二乘法的目标轨迹,有利于重新捕获目标。实验结果表明,与已有的跟踪方法相比,该方法在复杂背景和遮挡过程中对目标的定位更精确,鲁棒性更好。

关键词: 背景建模, 均值漂移算法, 多特征融合, 运动信息, 最小二乘法

Abstract: A novel target tracking method for target’s localization with mean shift based on multiple features and least squares method prediction is proposed to solve background interference,problems of partial occlusion and other issues.Aiming at the problems of background interference,the Improved Gaussian Mixture Model(IGMM) is proposed to build background model and to extracted foreground moving object. The motion information,color information and texture information are fused to represent the target,according to the change of the target and backgrounds,combined with the motion feature to remove noise interference of background and to locate the object. In the occlusion process,when the mean shift integrating multi-feature fails to track the target,it can use least squares method to predict the location of the target. Experimental results demonstrate that the method can track the target accurately and has better robustness in complex backgrounds or occlusion situations.

Key words: background modeling, mean shift algorithm, multi-feature fusion, motion information, least square method

中图分类号: